Journal of Modern Mathematics Frontier

JMMF

Using Extreme Value Theory to Optimally Design PV

Applications under Climate Change
Shinguang Chen
Department of Industrial Engineering and Management,
Tungnan University
New Taipei City, Taiwan
bobchen@mail.tnu.edu.tw
Abstract-A novel design method for photovoltaic applications
(PVA) is proposed based on Extreme Value Theory under climate
change. Traditionally, one day or one years climate data are
used in the design of PV applications. However, the fact is that
weather conditions in two years wont be the same.Therefore, the
PV applications designed by the traditional way may fail in the
extreme weather conditions. This paper illustrates a novel
method to design a PV application reliable under the climatechange situation.
Keywords-Climate Change; Extreme Value Theory; PV
Application; Sizing; Climate Cycle

I. INTRODUCTION
Design of Photovoltaic applications (PVA) is an active
research area in the literature. A very comprehensive review
about this topic can be referred to [1]. The technical
introduction of PVA can be found in [2]. The traditional
methods of power system reliability evaluation are discussed in
[3]. Gordon [4] gave an optimal sizing technique based on the
stochastic process. Egido and Lorenzo [5] presented an early
review and proposed a method to take the advantages from
both numerical and analytic approaches. Markvart et al. [6]
proposed a sizing technique based on the extreme events.
The design of PV applications mainly involves determining
the number of PV plates and the number of batteries in the
application. Most of them only use one day or one years data
to do the design task. However, the fact is that the weather
conditions in a single site in two years will not be the same.
Therefore, the PV applications designed by the traditional way
may fail in the extreme weather conditions.
Extreme Value Theory (EVT) [7, 8] is used in the modeling
of extreme events occurred in a time series. Since the maximal
climate cycle in a year normally cannot be predicted, it is taken
as a random event and can be modeled by EVT. EVT has
various findings that fit different situations. This paper mainly
illustrates the design of PV applications in terms of block
maximum. The realistic data from a real site are employed to
the illustration. A comprehensive version of this approach can
be referred to [9].The remainder of the work is described as
follows: The assumptions for illustrating the approach are
presented in Section 2. The mathematic preliminaries are
addressed in Section 3. The illustration of the proposed method
with realistic data is in Section 4. Finally, Section 5 is the
conclusion.
II. ASSUMPTIONS
To simplify the illustration, it is assumed to satisfy the
following assumptions:
1.

The system obeys energy conservation law.

2.

The load is constant.

3.

The system is optimized by the vendor and the average

power of the PV panel is used in calculation.

4.

The batteries are initially full.

5.

The batteries are only charged by the PV system at day.

6.

The load is supplied partly by the PV system at day and

totally by the batteries at night.
III.

PRELIMINARIES

A. Extreme Value Theory

EVT is a method used for the evaluation of LPPx under
observed data. The extreme value theory gives the following
important results [10]:
Theorem 3.1. Suppose X1, X2, , Xn are iid random
variables. If there are constants an R, bn > 0, Mn (the order
statistics for Xn) and some non-degenerate limit distribution H
such that
lim Pr(

M n an

x) H ( x)

bn

then H is one of the following extreme value distributions:

Frechet: ( x )

x0

0,

Weibull : ( x )

( x)

Gumbel : ( x ) e

x0

x0

x0

0
e

, 0,

, 0,

,xR

A more general form called the generalized extreme value

distribution (GEVD) of the three distributions can be expressed
as
H , ,

(1

x 1/
)

x u

0,
0,

where is the tail index, is the position parameter, is the

scaling parameter.
Theorem 3.1 states that the tail of any iid random variable
with unknown distribution will asymptotic converge to the
GEVD. This is a very important result of precisely estimating
LPPx mathematically.

JMMF Vol.1 No.1 2012 PP.6-8 www.sjmmf.org 2012 World Academic Publishing
6

Journal of Modern Mathematics Frontier

JMMF

B. Calculation of LPPx
LPPx is defined as:

D. Sizing Formulas

LPPx Pr( M n x ) 1 H , , (

Let Tcmax = max{Tci | 1 i Q}. How to choose a suitable

charging time T and get the corresponding Tcmax depends on the
given weather condition. Empirical study shows that the
relationship between Tcmax/T and T is not linear, actually it is a
nearly convex function. Figure 2 shows this relationship.

).

C. Climate Cycle Creation

The climate cycle can be defined as the specific charge time
to charge full the batteries. The following flow chart is
proposed to create such climate cycle:

Fig. 2 The convex function realtionship

Two sizing formulas are summarized from the observation.

For PV panels:
P0 L0 (1

to do the summation. By
We define a symbol
definition, a climate cycle ci may start from the si percentage of
the PSH at the last day of the previous cycle ci-1 and end at the
(1 - si+1) percentage of the PSH at the last day of this cycle. Let
Tci be the cycle time (i.e., the discharge time) of cycle ci, then it
is expressed as
si e( i 1) R ' 12

ci 24 Tci

e( i 1) R '

(1 si 1 )eiR 12

24( R 1)

eiR
2

1 i Q,

24( R 1) (1 si 1 )ei 12

eiR
2

, i 1.

In unit of a day, it is denoted as

e( i 1)R ' ( si 1 / 2) / 24 R eiR (1 / 2 si 1 ) / 24, 1 i Q,

Td i

R 1 / 2 ei (1 / 2 si 1 ) / 24,
R

i 1.

Tdi will be used in the later computation of LPPTdi.

Suppose there are Y years. According to the extreme value
theory for block maxima, we define the random variables as
Xj = max{Tdi|ci is in year j}, 1 j Y.

RTE

Tcmax
T

1)).

For batteries:

Fig. 1 The flowchart of creating climate cycles

VI

L0
RTE

Tcmax
T

1).

where RTE is round trip efficiency, P0 is the power of PV

panels, VI is the power of batteries.
IV. EXAMPLES
Two sets of the solar radiation data in Florida, USA are
illustrated: one is for the in-sample test measuring daily solar
radiation for 30 years from 1961 to 1990; the other is for the
out-of-sample test measuring daily solar radiation for 15 years
from 1991 to 2005. Both data are obtained from the Renewable
Resource Data Center of the National Renewable Energy
Laboratory, USA. Assume that there is a constant daily load of
4 kWh (48 V) uniformly distributed in a day to be powered by
the PV system (205 W, 48 V for one plate) with batteries
regulated, where 205W is the average power of the PV panels
during PSH. Then, each module of the battery bank needs 4
batteries (lead-acid type, 12 V, 100Ah) cascaded. The RTE of
the battery is 0.9.
At first, the empirical analysis of the ratio Tcmax/ T vs. T is
conducted. The minimal ratio 9.42851 is at 58.7 h (=T*) of the
turn in this analysis. The corresponding length of maximal
climate cycle is 23.06058 days Td *max.
In in-sample test, we got the optimal PV panels are 9 plates:

JMMF Vol.1 No.1 2012 PP.6-8 www.sjmmf.org 2012 World Academic Publishing
7

Journal of Modern Mathematics Frontier

The optimal batteries are 4 modules:

The reliability estimation is 0.02981 which comes from the

EVT fitting by MLE method. The fitting parameters are: =
0.000001, = 1.048 and = 20.947. Figure 3. is the fitted
curve.

JMMF

V. CONCLUSIONS
This paper illustrates the way of how to use the modern
mathematics in the real applications. A novel design method for
PV applications is proposed. Many interesting findings are
obtained, which are never explored before. More realistic
climate data are encouraged for the future verification of this
method.
ACKNOWLEDGMENT

This work was supported in part by the National Science

Council, Taiwan, Republic of China, under Grant No. NSC
100-2221-E-236-003.
REFERENCES
[1]

Fig. 3 The fitted extreme climate model

This result is as good as that by the traditional sizing curve

method, but with more mathematic insights.
In out-of-sample test, we verify what happened to this
design of PV applications. The climate statistics for the next 15
years are in Figure 4.

Fig. 4 The extreme events in the next 15 years

The results show that the optimal design is feasible for the
future climates. However, three extremes are discovered by this
method, which would occur at a frequency of per 317.5611,
48.146 and 47.7783 years respectively. These are the evidences
of climate change.

W. Zhou, C. Lou, Z. Li, L. Lu, and H. Yang, Current status of research

on optimum sizing of stand-alone hybrid solar-wind power generation
systems, Appl Energy, vol 87, pp. 380389, 2010.
[2] S. Silvestre, Review of system design and sizing tools, Oxford: Elsevier,
2003.
[3] R. Billinton, and R. N. Allan, Reliability evaluation of power systems,
2nd ed. New York and London: Plenum Press, 1992.
[4] J. M. Gordon, Optimal sizing of stand-alone photovoltaic solar power
systems, Sol Cells, vol 20, pp. 295313, 1987.
[5] M. Egido, and E. Lorenzo, The sizing of stand alone PV-system: a
review and a proposed new method, Sol Energy Mater Sol Cells, vol 26,
pp. 5169, 1992.
[6] T. Markvart, A. Fragaki, and J. N. Ross, PV system sizing using
observed time series of solar radiation, Sol Energy, vol 80, pp. 4650,
2006.
[7] E. Castillo, Extreme value theory in engineering. London/San Diego:
Academic Press, 1988.
[8] J. Galambos, Extreme value theory for applications, In: Galambos J,
Lechner J, Simiu E, editors. Proceedings of the conference on extreme
value theory and applications. Springer, pp. 1
14, 1993.
[9] S. G. Chen, An efficient sizing method for a
stand-alone PV system in terms of the observed
block extremes, Appl Energy, vol 91, pp. 375
384, 2012.
[10] R. A. Fisher, and L. H. C. Tippett, Limiting
forms of the frequency distribution of the largest
or smallest member of a sample, Math Proc
Cambridge Philos Soc, vol 24, pp. 18090,
1928.
Shin-Guang Chen received his BSc (1985) in Computer Engineering from
the National Chiao Tung University in Taiwan, and his PhD (1997) in
Industrial Engineering from the National Chiao Tung University in Taiwan.
Since 2001, he has been with the Department of Industrial Engineering &
Management at the Tungnan University in Taiwan, where he holds the rank of
Assistant Professor. His research interests are in management, reliability,
ERPS, and sustainable technologies. His recent work involves problems
related to the analysis of flow networks and renewable energy. He has
published papers in Omega, IJPR, IJIE, ESWA, JORSJ, CSTM, IJOR, JCIIE,
IJRQP, APEN, C&IE, etc.

JMMF Vol.1 No.1 2012 PP.6-8 www.sjmmf.org 2012 World Academic Publishing
8